skip to content
clipboard

Pessimistic vs Optimistic Locking

2 min read
Table of Contents

Overview

Locks are essential to maintain data consistency and integrity in multi-user environments. They prevent simultaneous modifications that can lead to data inconsistencies.


Pessimistic Locking

Assumption: Conflicts will occur → lock early

Aspect Detail
Behavior Locks data before any changes; prevents other users from reading/updating until lock released
Typical SQL SELECT ... FOR UPDATE (row-level), LOCK TABLE (table-level)
Isolation Often requires REPEATABLE READ or SERIALIZABLE
Best for High contention, short transactions, critical sections

Pros

  • Strong consistency guarantee
  • Simple mental model
  • Prevents conflicts entirely

Cons

  • Performance: Blocks other transactions
  • Deadlock risk: Circular wait conditions
  • Scalability: Limits concurrency

Optimistic Locking

Assumption: Conflicts are rare → check at commit

Aspect Detail
Behavior Allows concurrent reads/writes; validates at commit time (version/timestamp check)
Typical Implementation version column (increment on update), or updated_at timestamp
Conflict Handling Abort + retry on version mismatch
Best for Low contention, long transactions, high read-to-write ratio

Pros

  • High concurrency, no blocking
  • Better throughput under low contention
  • Works well with distributed systems

Cons

  • Retry logic complexity
  • Wasted work on conflict
  • Not suitable for high-contention hotspots

Comparison

Dimension Pessimistic Optimistic
Concurrency Low (blocks) High (non-blocking)
Consistency Immediate At commit
Deadlocks Possible None (but retries)
Latency Predictable Variable (retries)
Complexity Low Medium (retry logic)

Best Practices

  1. Hold locks minimum time — Short transactions
  2. Granular locks — Row-level > table-level
  3. Retry logic — Exponential backoff for optimistic conflicts
  4. Choose per use case:
    • Inventory decrement → Pessimistic
    • User profile update → Optimistic
    • Financial ledger → Pessimistic (or serialized)
    • Cache invalidation → Optimistic